lustre/quota/quota_interface.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_LQUOTA
  41
  42 #ifdef __KERNEL__
  43 # include <linux/version.h>
  44 # include <linux/module.h>
  45 # include <linux/init.h>
  46 # include <linux/fs.h>
  47 # include <linux/jbd.h>
  48 # if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
  49 #  include <linux/smp_lock.h>
  50 #  include <linux/buffer_head.h>
  51 #  include <linux/workqueue.h>
  52 #  include <linux/mount.h>
  53 # else
  54 #  include <linux/locks.h>
  55 # endif
  56 #else /* __KERNEL__ */
  57 # include <liblustre.h>
  58 #endif
  59
  60 #include <obd_class.h>
  61 #include <lustre_mds.h>
  62 #include <lustre_dlm.h>
  63 #include <lustre_cfg.h>
  64 #include <obd_ost.h>
  65 #include <lustre_fsfilt.h>
  66 #include <lustre_quota.h>
  67 #include <lprocfs_status.h>
  68 #include "quota_internal.h"
  69
  70 #ifdef __KERNEL__
  71
  72 #ifdef HAVE_QUOTA_SUPPORT
  73
  74 static cfs_time_t last_print = 0;
  75 static spinlock_t last_print_lock = SPIN_LOCK_UNLOCKED;
  76
  77 static int filter_quota_setup(struct obd_device *obd)
  78 {
  79         int rc = 0;
  80         struct obd_device_target *obt = &obd->u.obt;
  81         ENTRY;
  82
  83         init_rwsem(&obt->obt_rwsem);
  84         obt->obt_qfmt = LUSTRE_QUOTA_V2;
  85         atomic_set(&obt->obt_quotachecking, 1);
  86         rc = qctxt_init(obd, NULL);
  87         if (rc)
  88                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
  89
  90         RETURN(rc);
  91 }
  92
  93 static int filter_quota_cleanup(struct obd_device *obd)
  94 {
  95         ENTRY;
  96         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
  97         RETURN(0);
  98 }
  99
 100 static int filter_quota_setinfo(struct obd_device *obd, void *data)
 101 {
 102         struct obd_export *exp = data;
 103         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 104         struct obd_import *imp;
 105         ENTRY;
 106
 107         /* setup the quota context import */
 108         spin_lock(&qctxt->lqc_lock);
 109         qctxt->lqc_import = exp->exp_imp_reverse;
 110         spin_unlock(&qctxt->lqc_lock);
 111         CDEBUG(D_QUOTA, "%s: lqc_import(%p) of obd(%p) is reactivated now, \n",
 112                obd->obd_name,exp->exp_imp_reverse, obd);
 113
 114         /* make imp's connect flags equal relative exp's connect flags
 115          * adding it to avoid the scan export list
 116          */
 117         imp = qctxt->lqc_import;
 118         if (likely(imp))
 119                 imp->imp_connect_data.ocd_connect_flags |=
 120                         (exp->exp_connect_flags &
 121                          (OBD_CONNECT_QUOTA64 | OBD_CONNECT_CHANGE_QS));
 122
 123         cfs_waitq_signal(&qctxt->lqc_wait_for_qmaster);
 124         /* start quota slave recovery thread. (release high limits) */
 125         qslave_start_recovery(obd, qctxt);
 126         RETURN(0);
 127 }
 128
 129 static int filter_quota_clearinfo(struct obd_export *exp, struct obd_device *obd)
 130 {
 131         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 132         ENTRY;
 133
 134         /* lquota may be not set up before destroying export, b=14896 */
 135         if (!obd->obd_set_up)
 136                 RETURN(0);
 137
 138         /* when exp->exp_imp_reverse is destroyed, the corresponding lqc_import
 139          * should be invalid b=12374 */
 140         if (qctxt->lqc_import && qctxt->lqc_import == exp->exp_imp_reverse) {
 141                 spin_lock(&qctxt->lqc_lock);
 142                 qctxt->lqc_import = NULL;
 143                 spin_unlock(&qctxt->lqc_lock);
 144                 CDEBUG(D_QUOTA, "%s: lqc_import of obd(%p) is invalid now.\n",
 145                        obd->obd_name, obd);
 146         }
 147         RETURN(0);
 148 }
 149
 150 static int filter_quota_enforce(struct obd_device *obd, unsigned int ignore)
 151 {
 152         ENTRY;
 153
 154         if (!sb_any_quota_enabled(obd->u.obt.obt_sb))
 155                 RETURN(0);
 156
 157         if (ignore) {
 158                 CDEBUG(D_QUOTA, "blocks will be written with ignoring quota.\n");
 159                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
 160         } else {
 161                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
 162         }
 163
 164         RETURN(0);
 165 }
 166
 167 static int filter_quota_getflag(struct obd_device *obd, struct obdo *oa)
 168 {
 169         struct obd_device_target *obt = &obd->u.obt;
 170         struct lustre_quota_ctxt *qctxt = &obt->obt_qctxt;
 171         int err, cnt, rc = 0;
 172         struct obd_quotactl *oqctl;
 173         ENTRY;
 174
 175         if (!sb_any_quota_enabled(obt->obt_sb))
 176                 RETURN(0);
 177
 178         OBD_ALLOC_PTR(oqctl);
 179         if (!oqctl) {
 180                 CERROR("Not enough memory!");
 181                 RETURN(-ENOMEM);
 182         }
 183
 184         /* set over quota flags for a uid/gid */
 185         oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
 186         oa->o_flags &= ~(OBD_FL_NO_USRQUOTA | OBD_FL_NO_GRPQUOTA);
 187
 188         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 189                 struct quota_adjust_qunit oqaq_tmp;
 190                 struct lustre_qunit_size *lqs = NULL;
 191
 192                 oqaq_tmp.qaq_flags = cnt;
 193                 oqaq_tmp.qaq_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 194
 195                 quota_search_lqs(NULL, &oqaq_tmp, qctxt, &lqs);
 196                 if (lqs) {
 197                         spin_lock(&lqs->lqs_lock);
 198                         if (lqs->lqs_bunit_sz <= qctxt->lqc_sync_blk) {
 199                                 oa->o_flags |= (cnt == USRQUOTA) ?
 200                                         OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 201                                 spin_unlock(&lqs->lqs_lock);
 202                                 CDEBUG(D_QUOTA, "set sync flag: bunit(%lu), "
 203                                        "sync_blk(%d)\n", lqs->lqs_bunit_sz,
 204                                        qctxt->lqc_sync_blk);
 205                                 /* this is for quota_search_lqs */
 206                                 lqs_putref(lqs);
 207                                 continue;
 208                         }
 209                         spin_unlock(&lqs->lqs_lock);
 210                         /* this is for quota_search_lqs */
 211                         lqs_putref(lqs);
 212                 }
 213
 214                 memset(oqctl, 0, sizeof(*oqctl));
 215
 216                 oqctl->qc_cmd = Q_GETQUOTA;
 217                 oqctl->qc_type = cnt;
 218                 oqctl->qc_id = (cnt == USRQUOTA) ? oa->o_uid : oa->o_gid;
 219                 err = fsfilt_quotactl(obd, obt->obt_sb, oqctl);
 220                 if (err) {
 221                         if (!rc)
 222                                 rc = err;
 223                         oa->o_valid &= ~((cnt == USRQUOTA) ? OBD_MD_FLUSRQUOTA :
 224                                                              OBD_MD_FLGRPQUOTA);
 225                         continue;
 226                 }
 227
 228                 if (oqctl->qc_dqblk.dqb_bhardlimit &&
 229                    (toqb(oqctl->qc_dqblk.dqb_curspace) >=
 230                     oqctl->qc_dqblk.dqb_bhardlimit))
 231                         oa->o_flags |= (cnt == USRQUOTA) ?
 232                                 OBD_FL_NO_USRQUOTA : OBD_FL_NO_GRPQUOTA;
 233         }
 234         OBD_FREE_PTR(oqctl);
 235         RETURN(rc);
 236 }
 237
 238 /**
 239  * check whether the left quota of certain uid and gid can satisfy a block_write
 240  * or inode_create rpc. When need to acquire quota, return QUOTA_RET_ACQUOTA
 241  */
 242 static int quota_check_common(struct obd_device *obd, unsigned int uid,
 243                               unsigned int gid, int count, int cycle, int isblk)
 244 {
 245         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 246         int i;
 247         __u32 id[MAXQUOTAS] = { uid, gid };
 248         struct qunit_data qdata[MAXQUOTAS];
 249         int rc = 0, rc2[2] = { 0, 0 };
 250         ENTRY;
 251
 252         CLASSERT(MAXQUOTAS < 4);
 253         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 254                 RETURN(rc);
 255
 256         spin_lock(&qctxt->lqc_lock);
 257         if (!qctxt->lqc_valid){
 258                 spin_unlock(&qctxt->lqc_lock);
 259                 RETURN(rc);
 260         }
 261         spin_unlock(&qctxt->lqc_lock);
 262
 263         for (i = 0; i < MAXQUOTAS; i++) {
 264                 struct lustre_qunit_size *lqs = NULL;
 265
 266                 qdata[i].qd_id = id[i];
 267                 qdata[i].qd_flags = i;
 268                 if (isblk)
 269                         QDATA_SET_BLK(&qdata[i]);
 270                 qdata[i].qd_count = 0;
 271
 272                 /* ignore root user */
 273                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 274                         continue;
 275
 276                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 277                 if (!lqs)
 278                         continue;
 279
 280                 rc2[i] = compute_remquota(obd, qctxt, &qdata[i], isblk);
 281                 spin_lock(&lqs->lqs_lock);
 282                 if (!cycle) {
 283                         rc = QUOTA_RET_INC_PENDING;
 284                         if (isblk)
 285                                 lqs->lqs_bwrite_pending += count;
 286                         else
 287                                 lqs->lqs_iwrite_pending += count;
 288                 }
 289                 if (rc2[i] == QUOTA_RET_OK) {
 290                         if (isblk && qdata[i].qd_count <
 291                             lqs->lqs_bwrite_pending * CFS_PAGE_SIZE)
 292                                 rc2[i] = QUOTA_RET_ACQUOTA;
 293                         if (!isblk && qdata[i].qd_count <
 294                             lqs->lqs_iwrite_pending)
 295                                 rc2[i] = QUOTA_RET_ACQUOTA;
 296                 }
 297                 spin_unlock(&lqs->lqs_lock);
 298                 CDEBUG(D_QUOTA, "count: %d, write pending: %lu, qd_count: "LPU64
 299                        ".\n", count,
 300                        isblk ? lqs->lqs_bwrite_pending : lqs->lqs_iwrite_pending,
 301                        qdata[i].qd_count);
 302
 303                 /* When cycle is zero, lqs_*_pending will be changed. We will
 304                  * get reference of the lqs here and put reference of lqs in
 305                  * quota_pending_commit b=14784 */
 306                 if (!cycle)
 307                         lqs_getref(lqs);
 308
 309                 /* this is for quota_search_lqs */
 310                 lqs_putref(lqs);
 311         }
 312
 313         if (rc2[0] == QUOTA_RET_ACQUOTA || rc2[1] == QUOTA_RET_ACQUOTA)
 314                 RETURN(rc | QUOTA_RET_ACQUOTA);
 315         else
 316                 RETURN(rc);
 317 }
 318
 319 static int quota_chk_acq_common(struct obd_device *obd, unsigned int uid,
 320                                 unsigned int gid, int count, int *pending,
 321                                 quota_acquire acquire,
 322                                 struct obd_trans_info *oti, int isblk)
 323 {
 324         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 325         struct timeval work_start;
 326         struct timeval work_end;
 327         long timediff;
 328         struct l_wait_info lwi = { 0 };
 329         int rc = 0, cycle = 0, count_err = 1;
 330         ENTRY;
 331
 332         CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
 333         *pending = 0;
 334         /* Unfortunately, if quota master is too busy to handle the
 335          * pre-dqacq in time and quota hash on ost is used up, we
 336          * have to wait for the completion of in flight dqacq/dqrel,
 337          * in order to get enough quota for write b=12588 */
 338         do_gettimeofday(&work_start);
 339         while ((rc = quota_check_common(obd, uid, gid, count, cycle, isblk)) &
 340                QUOTA_RET_ACQUOTA) {
 341
 342                 spin_lock(&qctxt->lqc_lock);
 343                 if (!qctxt->lqc_import && oti) {
 344                         spin_unlock(&qctxt->lqc_lock);
 345
 346                         LASSERT(oti && oti->oti_thread &&
 347                                 oti->oti_thread->t_watchdog);
 348
 349                         lc_watchdog_disable(oti->oti_thread->t_watchdog);
 350                         CDEBUG(D_QUOTA, "sleep for quota master\n");
 351                         l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
 352                                      &lwi);
 353                         CDEBUG(D_QUOTA, "wake up when quota master is back\n");
 354                         lc_watchdog_touch(oti->oti_thread->t_watchdog);
 355                 } else {
 356                         spin_unlock(&qctxt->lqc_lock);
 357                 }
 358
 359                 if (rc & QUOTA_RET_INC_PENDING)
 360                         *pending = 1;
 361
 362                 cycle++;
 363                 if (isblk)
 364                         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
 365                 /* after acquire(), we should run quota_check_common again
 366                  * so that we confirm there are enough quota to finish write */
 367                 rc = acquire(obd, uid, gid, oti, isblk);
 368
 369                 /* please reference to dqacq_completion for the below */
 370                 /* a new request is finished, try again */
 371                 if (rc == -EAGAIN) {
 372                         CDEBUG(D_QUOTA, "finish a quota req, try again\n");
 373                         continue;
 374                 }
 375
 376                 /* it is out of quota already */
 377                 if (rc == -EDQUOT) {
 378                         CDEBUG(D_QUOTA, "out of quota,  return -EDQUOT\n");
 379                         break;
 380                 }
 381
 382                 /* -EBUSY and others, wait a second and try again */
 383                 if (rc < 0) {
 384                         cfs_waitq_t        waitq;
 385                         struct l_wait_info lwi;
 386
 387                         if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
 388                                 lc_watchdog_touch(oti->oti_thread->t_watchdog);
 389                         CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
 390                                count_err++);
 391
 392                         init_waitqueue_head(&waitq);
 393                         lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
 394                                           NULL);
 395                         l_wait_event(waitq, 0, &lwi);
 396                 }
 397
 398                 if (rc < 0 || cycle % 10 == 2) {
 399                         spin_lock(&last_print_lock);
 400                         if (last_print == 0 ||
 401                             cfs_time_before((last_print + cfs_time_seconds(30)),
 402                                             cfs_time_current())) {
 403                                 last_print = cfs_time_current();
 404                                 spin_unlock(&last_print_lock);
 405                                 CWARN("still haven't managed to acquire quota "
 406                                       "space from the quota master after %d "
 407                                       "retries (err=%d, rc=%d)\n",
 408                                       cycle, count_err - 1, rc);
 409                         } else {
 410                                 spin_unlock(&last_print_lock);
 411                         }
 412                 }
 413
 414                 CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
 415                        cycle);
 416         }
 417
 418         if (!cycle && rc & QUOTA_RET_INC_PENDING)
 419                 *pending = 1;
 420
 421         do_gettimeofday(&work_end);
 422         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 423         lprocfs_counter_add(qctxt->lqc_stats,
 424                             isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
 425                                     LQUOTA_WAIT_FOR_CHK_INO,
 426                             timediff);
 427
 428         RETURN(rc);
 429 }
 430
 431 /**
 432  * when a block_write or inode_create rpc is finished, adjust the record for
 433  * pending blocks and inodes
 434  */
 435 static int quota_pending_commit(struct obd_device *obd, unsigned int uid,
 436                                 unsigned int gid, int count, int isblk)
 437 {
 438         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 439         struct timeval work_start;
 440         struct timeval work_end;
 441         long timediff;
 442         int i;
 443         __u32 id[MAXQUOTAS] = { uid, gid };
 444         struct qunit_data qdata[MAXQUOTAS];
 445         ENTRY;
 446
 447         CDEBUG(D_QUOTA, "commit pending quota for  %s\n", obd->obd_name);
 448         CLASSERT(MAXQUOTAS < 4);
 449         if (!sb_any_quota_enabled(qctxt->lqc_sb))
 450                 RETURN(0);
 451
 452         do_gettimeofday(&work_start);
 453         for (i = 0; i < MAXQUOTAS; i++) {
 454                 struct lustre_qunit_size *lqs = NULL;
 455
 456                 qdata[i].qd_id = id[i];
 457                 qdata[i].qd_flags = i;
 458                 if (isblk)
 459                         QDATA_SET_BLK(&qdata[i]);
 460                 qdata[i].qd_count = 0;
 461
 462                 if (qdata[i].qd_id == 0 && !QDATA_IS_GRP(&qdata[i]))
 463                         continue;
 464
 465                 quota_search_lqs(&qdata[i], NULL, qctxt, &lqs);
 466                 if (lqs) {
 467                         int flag = 0;
 468                         CDEBUG(D_QUOTA, "pending: %lu, count: %d.\n",
 469                                isblk ? lqs->lqs_bwrite_pending :
 470                                lqs->lqs_iwrite_pending, count);
 471                         spin_lock(&lqs->lqs_lock);
 472                         if (isblk) {
 473                                 if (lqs->lqs_bwrite_pending >= count) {
 474                                         lqs->lqs_bwrite_pending -= count;
 475                                         spin_unlock(&lqs->lqs_lock);
 476                                         flag = 1;
 477                                 } else {
 478                                         spin_unlock(&lqs->lqs_lock);
 479                                         CDEBUG(D_ERROR,
 480                                                "there are too many blocks!\n");
 481                                 }
 482                         } else {
 483                                 if (lqs->lqs_iwrite_pending >= count) {
 484                                         lqs->lqs_iwrite_pending -= count;
 485                                         spin_unlock(&lqs->lqs_lock);
 486                                         flag = 1;
 487                                 } else {
 488                                         spin_unlock(&lqs->lqs_lock);
 489                                         CDEBUG(D_ERROR,
 490                                                "there are too many files!\n");
 491                                 }
 492                         }
 493
 494                         lqs_putref(lqs);
 495                         /* When lqs_*_pening is changed back, we'll putref lqs
 496                          * here b=14784 */
 497                         if (flag)
 498                                 lqs_putref(lqs);
 499                 }
 500         }
 501         do_gettimeofday(&work_end);
 502         timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
 503         lprocfs_counter_add(qctxt->lqc_stats,
 504                             isblk ? LQUOTA_WAIT_FOR_COMMIT_BLK :
 505                                     LQUOTA_WAIT_FOR_COMMIT_INO,
 506                             timediff);
 507
 508         RETURN(0);
 509 }
 510
 511 static int mds_quota_init(void)
 512 {
 513         return lustre_dquot_init();
 514 }
 515
 516 static int mds_quota_exit(void)
 517 {
 518         lustre_dquot_exit();
 519         return 0;
 520 }
 521
 522 static int mds_quota_setup(struct obd_device *obd)
 523 {
 524         struct obd_device_target *obt = &obd->u.obt;
 525         struct mds_obd *mds = &obd->u.mds;
 526         int rc;
 527         ENTRY;
 528
 529         if (unlikely(mds->mds_quota)) {
 530                 CWARN("try to reinitialize quota context!\n");
 531                 RETURN(0);
 532         }
 533
 534         init_rwsem(&obt->obt_rwsem);
 535         obt->obt_qfmt = LUSTRE_QUOTA_V2;
 536         mds->mds_quota_info.qi_version = LUSTRE_QUOTA_V2;
 537         atomic_set(&obt->obt_quotachecking, 1);
 538         /* initialize quota master and quota context */
 539         sema_init(&mds->mds_qonoff_sem, 1);
 540         rc = qctxt_init(obd, dqacq_handler);
 541         if (rc) {
 542                 CERROR("initialize quota context failed! (rc:%d)\n", rc);
 543                 RETURN(rc);
 544         }
 545         mds->mds_quota = 1;
 546         RETURN(rc);
 547 }
 548
 549 static int mds_quota_cleanup(struct obd_device *obd)
 550 {
 551         ENTRY;
 552         if (unlikely(!obd->u.mds.mds_quota))
 553                 RETURN(0);
 554
 555         qctxt_cleanup(&obd->u.obt.obt_qctxt, 0);
 556         RETURN(0);
 557 }
 558
 559 static int mds_quota_setinfo(struct obd_device *obd, void *data)
 560 {
 561         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 562         ENTRY;
 563
 564         if (unlikely(!obd->u.mds.mds_quota))
 565                 RETURN(0);
 566
 567         if (data != NULL)
 568                 QUOTA_MASTER_READY(qctxt);
 569         else
 570                 QUOTA_MASTER_UNREADY(qctxt);
 571         RETURN(0);
 572 }
 573
 574 static int mds_quota_fs_cleanup(struct obd_device *obd)
 575 {
 576         struct mds_obd *mds = &obd->u.mds;
 577         struct obd_quotactl oqctl;
 578         ENTRY;
 579
 580         if (unlikely(!mds->mds_quota))
 581                 RETURN(0);
 582
 583         mds->mds_quota = 0;
 584         memset(&oqctl, 0, sizeof(oqctl));
 585         oqctl.qc_type = UGQUOTA;
 586
 587         down(&mds->mds_qonoff_sem);
 588         mds_admin_quota_off(obd, &oqctl);
 589         up(&mds->mds_qonoff_sem);
 590         RETURN(0);
 591 }
 592
 593 static int quota_acquire_common(struct obd_device *obd, unsigned int uid,
 594                                 unsigned int gid, struct obd_trans_info *oti,
 595                                 int isblk)
 596 {
 597         struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
 598         int rc;
 599         ENTRY;
 600
 601         rc = qctxt_adjust_qunit(obd, qctxt, uid, gid, isblk, 1, oti);
 602         RETURN(rc);
 603 }
 604
 605 #endif /* HAVE_QUOTA_SUPPORT */
 606 #endif /* __KERNEL__ */
 607
 608 struct osc_quota_info {
 609         struct list_head        oqi_hash;       /* hash list */
 610         struct client_obd      *oqi_cli;        /* osc obd */
 611         unsigned int            oqi_id;         /* uid/gid of a file */
 612         short                   oqi_type;       /* quota type */
 613 };
 614
 615 spinlock_t qinfo_list_lock = SPIN_LOCK_UNLOCKED;
 616
 617 static struct list_head qinfo_hash[NR_DQHASH];
 618 /* SLAB cache for client quota context */
 619 cfs_mem_cache_t *qinfo_cachep = NULL;
 620
 621 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 622                          __attribute__((__const__));
 623
 624 static inline int hashfn(struct client_obd *cli, unsigned long id, int type)
 625 {
 626         unsigned long tmp = ((unsigned long)cli>>6) ^ id;
 627         tmp = (tmp * (MAXQUOTAS - type)) % NR_DQHASH;
 628         return tmp;
 629 }
 630
 631 /* caller must hold qinfo_list_lock */
 632 static inline void insert_qinfo_hash(struct osc_quota_info *oqi)
 633 {
 634         struct list_head *head = qinfo_hash +
 635                 hashfn(oqi->oqi_cli, oqi->oqi_id, oqi->oqi_type);
 636
 637         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 638         list_add(&oqi->oqi_hash, head);
 639 }
 640
 641 /* caller must hold qinfo_list_lock */
 642 static inline void remove_qinfo_hash(struct osc_quota_info *oqi)
 643 {
 644         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 645         list_del_init(&oqi->oqi_hash);
 646 }
 647
 648 /* caller must hold qinfo_list_lock */
 649 static inline struct osc_quota_info *find_qinfo(struct client_obd *cli,
 650                                                 unsigned int id, int type)
 651 {
 652         unsigned int hashent = hashfn(cli, id, type);
 653         struct osc_quota_info *oqi;
 654         ENTRY;
 655
 656         LASSERT_SPIN_LOCKED(&qinfo_list_lock);
 657         list_for_each_entry(oqi, &qinfo_hash[hashent], oqi_hash) {
 658                 if (oqi->oqi_cli == cli &&
 659                     oqi->oqi_id == id && oqi->oqi_type == type)
 660                         return oqi;
 661         }
 662         RETURN(NULL);
 663 }
 664
 665 static struct osc_quota_info *alloc_qinfo(struct client_obd *cli,
 666                                           unsigned int id, int type)
 667 {
 668         struct osc_quota_info *oqi;
 669         ENTRY;
 670
 671         OBD_SLAB_ALLOC(oqi, qinfo_cachep, CFS_ALLOC_STD, sizeof(*oqi));
 672         if(!oqi)
 673                 RETURN(NULL);
 674
 675         CFS_INIT_LIST_HEAD(&oqi->oqi_hash);
 676         oqi->oqi_cli = cli;
 677         oqi->oqi_id = id;
 678         oqi->oqi_type = type;
 679
 680         RETURN(oqi);
 681 }
 682
 683 static void free_qinfo(struct osc_quota_info *oqi)
 684 {
 685         OBD_SLAB_FREE(oqi, qinfo_cachep, sizeof(*oqi));
 686 }
 687
 688 int osc_quota_chkdq(struct client_obd *cli, unsigned int uid, unsigned int gid)
 689 {
 690         unsigned int id;
 691         int cnt, rc = QUOTA_OK;
 692         ENTRY;
 693
 694         spin_lock(&qinfo_list_lock);
 695         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 696                 struct osc_quota_info *oqi = NULL;
 697
 698                 id = (cnt == USRQUOTA) ? uid : gid;
 699                 oqi = find_qinfo(cli, id, cnt);
 700                 if (oqi) {
 701                         rc = NO_QUOTA;
 702                         break;
 703                 }
 704         }
 705         spin_unlock(&qinfo_list_lock);
 706
 707         RETURN(rc);
 708 }
 709
 710 int osc_quota_setdq(struct client_obd *cli, unsigned int uid, unsigned int gid,
 711                     obd_flag valid, obd_flag flags)
 712 {
 713         unsigned int id;
 714         obd_flag noquota;
 715         int cnt, rc = 0;
 716         ENTRY;
 717
 718
 719         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
 720                 struct osc_quota_info *oqi, *old;
 721
 722                 if (!(valid & ((cnt == USRQUOTA) ?
 723                     OBD_MD_FLUSRQUOTA : OBD_MD_FLGRPQUOTA)))
 724                         continue;
 725
 726                 id = (cnt == USRQUOTA) ? uid : gid;
 727                 noquota = (cnt == USRQUOTA) ?
 728                     (flags & OBD_FL_NO_USRQUOTA) : (flags & OBD_FL_NO_GRPQUOTA);
 729
 730                 oqi = alloc_qinfo(cli, id, cnt);
 731                 if (oqi) {
 732                         spin_lock(&qinfo_list_lock);
 733
 734                         old = find_qinfo(cli, id, cnt);
 735                         if (old && !noquota)
 736                                 remove_qinfo_hash(old);
 737                         else if (!old && noquota)
 738                                 insert_qinfo_hash(oqi);
 739
 740                         spin_unlock(&qinfo_list_lock);
 741
 742                         if (old || !noquota)
 743                                 free_qinfo(oqi);
 744                         if (old && !noquota)
 745                                 free_qinfo(old);
 746                 } else {
 747                         CERROR("not enough mem!\n");
 748                         rc = -ENOMEM;
 749                         break;
 750                 }
 751         }
 752
 753         RETURN(rc);
 754 }
 755
 756 int osc_quota_cleanup(struct obd_device *obd)
 757 {
 758         struct client_obd *cli = &obd->u.cli;
 759         struct osc_quota_info *oqi, *n;
 760         int i;
 761         ENTRY;
 762
 763         spin_lock(&qinfo_list_lock);
 764         for (i = 0; i < NR_DQHASH; i++) {
 765                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 766                         if (oqi->oqi_cli != cli)
 767                                 continue;
 768                         remove_qinfo_hash(oqi);
 769                         free_qinfo(oqi);
 770                 }
 771         }
 772         spin_unlock(&qinfo_list_lock);
 773
 774         RETURN(0);
 775 }
 776
 777 int osc_quota_init(void)
 778 {
 779         int i;
 780         ENTRY;
 781
 782         LASSERT(qinfo_cachep == NULL);
 783         qinfo_cachep = cfs_mem_cache_create("osc_quota_info",
 784                                             sizeof(struct osc_quota_info),
 785                                             0, 0);
 786         if (!qinfo_cachep)
 787                 RETURN(-ENOMEM);
 788
 789         for (i = 0; i < NR_DQHASH; i++)
 790                 CFS_INIT_LIST_HEAD(qinfo_hash + i);
 791
 792         RETURN(0);
 793 }
 794
 795 int osc_quota_exit(void)
 796 {
 797         struct osc_quota_info *oqi, *n;
 798         int i, rc;
 799         ENTRY;
 800
 801         spin_lock(&qinfo_list_lock);
 802         for (i = 0; i < NR_DQHASH; i++) {
 803                 list_for_each_entry_safe(oqi, n, &qinfo_hash[i], oqi_hash) {
 804                         remove_qinfo_hash(oqi);
 805                         free_qinfo(oqi);
 806                 }
 807         }
 808         spin_unlock(&qinfo_list_lock);
 809
 810         rc = cfs_mem_cache_destroy(qinfo_cachep);
 811         LASSERTF(rc == 0, "couldn't destory qinfo_cachep slab\n");
 812         qinfo_cachep = NULL;
 813
 814         RETURN(0);
 815 }
 816
 817 #ifdef __KERNEL__
 818 #ifdef HAVE_QUOTA_SUPPORT
 819 quota_interface_t mds_quota_interface = {
 820         .quota_init     = mds_quota_init,
 821         .quota_exit     = mds_quota_exit,
 822         .quota_setup    = mds_quota_setup,
 823         .quota_cleanup  = mds_quota_cleanup,
 824         .quota_check    = target_quota_check,
 825         .quota_ctl      = mds_quota_ctl,
 826         .quota_setinfo  = mds_quota_setinfo,
 827         .quota_fs_cleanup = mds_quota_fs_cleanup,
 828         .quota_recovery = mds_quota_recovery,
 829         .quota_adjust   = mds_quota_adjust,
 830         .quota_chkquota = quota_chk_acq_common,
 831         .quota_acquire  = quota_acquire_common,
 832         .quota_pending_commit = quota_pending_commit,
 833 };
 834
 835 quota_interface_t filter_quota_interface = {
 836         .quota_setup    = filter_quota_setup,
 837         .quota_cleanup  = filter_quota_cleanup,
 838         .quota_check    = target_quota_check,
 839         .quota_ctl      = filter_quota_ctl,
 840         .quota_setinfo  = filter_quota_setinfo,
 841         .quota_clearinfo = filter_quota_clearinfo,
 842         .quota_enforce  = filter_quota_enforce,
 843         .quota_getflag  = filter_quota_getflag,
 844         .quota_acquire  = quota_acquire_common,
 845         .quota_adjust   = filter_quota_adjust,
 846         .quota_chkquota = quota_chk_acq_common,
 847         .quota_adjust_qunit   = filter_quota_adjust_qunit,
 848         .quota_pending_commit = quota_pending_commit,
 849 };
 850 #endif
 851 #endif /* __KERNEL__ */
 852
 853 quota_interface_t mdc_quota_interface = {
 854         .quota_ctl      = client_quota_ctl,
 855         .quota_check    = client_quota_check,
 856         .quota_poll_check = client_quota_poll_check,
 857 };
 858
 859 quota_interface_t lmv_quota_interface = {
 860         .quota_ctl      = lmv_quota_ctl,
 861         .quota_check    = lmv_quota_check,
 862 };
 863
 864 quota_interface_t osc_quota_interface = {
 865         .quota_ctl      = client_quota_ctl,
 866         .quota_check    = client_quota_check,
 867         .quota_poll_check = client_quota_poll_check,
 868         .quota_init     = osc_quota_init,
 869         .quota_exit     = osc_quota_exit,
 870         .quota_chkdq    = osc_quota_chkdq,
 871         .quota_setdq    = osc_quota_setdq,
 872         .quota_cleanup  = osc_quota_cleanup,
 873         .quota_adjust_qunit = client_quota_adjust_qunit,
 874 };
 875
 876 quota_interface_t lov_quota_interface = {
 877         .quota_ctl      = lov_quota_ctl,
 878         .quota_check    = lov_quota_check,
 879         .quota_adjust_qunit = lov_quota_adjust_qunit,
 880 };
 881
 882 #ifdef __KERNEL__
 883
 884 cfs_proc_dir_entry_t *lquota_type_proc_dir = NULL;
 885
 886 static int __init init_lustre_quota(void)
 887 {
 888 #ifdef HAVE_QUOTA_SUPPORT
 889         int rc = 0;
 890
 891         lquota_type_proc_dir = lprocfs_register(OBD_LQUOTA_DEVICENAME,
 892                                                 proc_lustre_root,
 893                                                 NULL, NULL);
 894         if (IS_ERR(lquota_type_proc_dir)) {
 895                 CERROR("LProcFS failed in lquota-init\n");
 896                 rc = PTR_ERR(lquota_type_proc_dir);
 897                 return rc;
 898         }
 899
 900         rc = qunit_cache_init();
 901         if (rc)
 902                 return rc;
 903
 904         PORTAL_SYMBOL_REGISTER(filter_quota_interface);
 905         PORTAL_SYMBOL_REGISTER(mds_quota_interface);
 906 #endif
 907         PORTAL_SYMBOL_REGISTER(mdc_quota_interface);
 908         PORTAL_SYMBOL_REGISTER(lmv_quota_interface);
 909         PORTAL_SYMBOL_REGISTER(osc_quota_interface);
 910         PORTAL_SYMBOL_REGISTER(lov_quota_interface);
 911         return 0;
 912 }
 913
 914 static void /*__exit*/ exit_lustre_quota(void)
 915 {
 916         PORTAL_SYMBOL_UNREGISTER(mdc_quota_interface);
 917         PORTAL_SYMBOL_UNREGISTER(lmv_quota_interface);
 918         PORTAL_SYMBOL_UNREGISTER(osc_quota_interface);
 919         PORTAL_SYMBOL_UNREGISTER(lov_quota_interface);
 920 #ifdef HAVE_QUOTA_SUPPORT
 921         PORTAL_SYMBOL_UNREGISTER(filter_quota_interface);
 922         PORTAL_SYMBOL_UNREGISTER(mds_quota_interface);
 923
 924         qunit_cache_cleanup();
 925
 926         if (lquota_type_proc_dir)
 927                 lprocfs_remove(&lquota_type_proc_dir);
 928 #endif
 929 }
 930
 931 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 932 MODULE_DESCRIPTION("Lustre Quota");
 933 MODULE_LICENSE("GPL");
 934
 935 cfs_module(lquota, "1.0.0", init_lustre_quota, exit_lustre_quota);
 936
 937 #ifdef HAVE_QUOTA_SUPPORT
 938 EXPORT_SYMBOL(mds_quota_interface);
 939 EXPORT_SYMBOL(filter_quota_interface);
 940 #endif
 941 EXPORT_SYMBOL(mdc_quota_interface);
 942 EXPORT_SYMBOL(lmv_quota_interface);
 943 EXPORT_SYMBOL(osc_quota_interface);
 944 EXPORT_SYMBOL(lov_quota_interface);
 945 #endif /* __KERNEL */